Re: [PATCH v11 09/10] genirq/msi: map/unmap the MSI doorbells on msi_domain_alloc/free_irqs

[Thomas Gleixner]

On Tue, 26 Jul 2016, Auger Eric wrote:
> On 26/07/2016 11:00, Thomas Gleixner wrote:
> > In your case you don't want to have a partial allocation, so instead of
> > playing silly games with desc->irq you should add a flag which tells the PCI
> > code that you are not interested in a partial allocation and that it should
> > return an error code instead.
> 

> In that case can we consider we even succeeded in allocating 1 MSI? In case
> the IOMMU mapping fails, the MSI transaction will never reach the target MSI
> frame so it is not usable. So when you mean "partial" I understand we did
> not succeed in allocating maxvec IRQs, correct? Here we succeeded in
> allocating 0 IRQ and still msi_capability_init returns 1.
>
> msi_capability_init doc-comment says "a positive return value indicates the
> number of interrupts which could have been allocated."
> 
> I understand allocation success currently only depends on the fact virq was
> allocated and set to desc->irq. But with that IOMMU stuff doesn't the
> criteria changes?

Right. But then you need to express it differently in a consistent way. Not by
hacking around it by setting desc->irq to 0.

Something like a flag field in msi_desc which denotes various properties would
be a possible solution. MSI_IRQ_ALLOCATED and MSI_IRQ_REMAPPED would be
sufficient for now. And the deallocation/cleanup would rely on those flags
rather than checking desc->irq.

Thanks,

	tglx